CT 2.5.2
Visualizing and Analyzing Networks of Co-Purchased Books, CDs and DVDs
T. Iba1 2 and M. Mori3
1
Faculty of Policy Management, Keio University, Endo 5322, Fujisawa, Kanagawa, Japan
3
Research Fellow, Rakuten Institute of Technology Lab., Shinagawa Seaside Rakuten Tower, 4-12-3
Higashi-shinagawa, Shinagawa-ku, Tokyo, Japan
3
Rakuten Institute of Technology Lab., Shinagawa Seaside Rakuten Tower, 4-12-3 Higashi-shinagawa,
Shinagawa-ku, Tokyo, Japan
Introduction
Every day, many customers buy many kinds of products like books, CD, and DVD etc,
at online stores. Then, the companies often have giant data stock of the transactions.
In academic viewpoint, the data is important because it is able to become a clue to
understand the complexity of the market. The main issue is what kind of order is
emerged as a result of compiling the customers’ actions? For the purpose, we propose
the method to investigate the co-purchase network of the market, and also visualize
and analyze the network with using the real market data of Books, CDs, and DVDs of
the online store “Rakuten Books” (http://books.rakuten. co.jp/), which is one of the
biggest online stores in Japan. Note that this research was done as an analysis by
Rakuten Institute of Technology, and the data do not include any personal information.
Method
The co-purchase network is compiled by the following way. We describe a node A if
there is the product A is purchased by the target customers. Then we describe an edge
to connect node A and node B if the product A and the product B is purchased by a
customer. For describing the edge, we try two types of connection method: “full
connection” and “sequential connection” (Figure 1). In the former method, all the
nodes which user bought connect each other. In the latter method, nodes connect as
the sequential order of user bought. It means that an undirected graph is generated by
the former method and a direct graph is generated by the latter.
Figure 1: Two Connection Methods
Results
We visualize the map as a network of the relation among products based on choices
by customers. In the case of “full connection” with threshold to visualize the link, we
can understand that there are los of components stands for the hidden relationship of
the products (Figure 2). We also found the rank distribution of link weight follows
power-law in both case of the full-connection and sequential-connection method
(Figure 2 and 3).
Figure 2: Co-purchase Networks of Books, CDs, DVDs and Rank distributions of Link-
weight (Full-Connection, visualizing links more than weight 2)
Figure 3: Co-purchase Networks of Books, CDs, DVDs and Rank distributions of Link-
weight (Sequential-Connection)
Acknowledgement
We thank the project members: R. Nishida, S. Itoh, Y. Kitayama, and
M. Yoshida for the discussion, visualization and analysis. We also
thank the members of Rakuten Institute of Technology.
References
Y. Kitayama, M. Yoshida, S. Takami and T. Iba (2008): Analyzing Co-Purchase Network of Books in
Japanese Online Store, poster, International Conference of Network Science '08 (submitted)
R. Nishida, M. Mori and T. Iba (2008): Analyzing Co-Purchase Network of CDs in Japanese Online
Store, poster, International Conference of Network Science '08 (submitted)
S. Itoh, S. Takami and T. Iba (2008): Analyzing Co-Purchase Network of DVDs in Japanese Online
Store, poster, International Conference of Network Science '08 (submitted)